We suggest a performance attribution model which is adapted to the requirements of hedge funds by a set of style-consistent benchmark indices obtained from a neural network based clustering procedure. We compare our approach with alternative models and analyze whether fund of hedge funds managers create added value. Employing factor loadingchanges as a proxy for style changes, we show that style shifts are a characteristic feature of hedge funds and that style-consistency does not generally ameliorate performance. Finally, we demonstrate that while poor performers which change style can expect a subsequent performance improvement, the same does not hold for top performers.